Ceyao Zhang

One Demonstration Is Enough for Real-World Robotic Reinforcement Learning

Yuwan Liu, Hongze Yu, Song Liu, Yuhan Wang, Junge Zhang, Yaodong Yang, et al.

Learning effective robot control policies on physical hardware is challenging due to costly data collection and the difficulty of reward specification. Prior work has incorporated demonstrations into reinforcement learning (RL), yet existing approaches either require large number…Read more